Tokens tell you how much you paid. Tasks tell you what you got.
Task-level efficiency analytics for OpenClaw. Stop counting tokens — start measuring what actually got done and what it cost.
Tell OpenClaw: "Install the openclaw-tally skill." The agent will handle the installation and configuration automatically.
If you prefer the terminal, run:
clawhub install openclaw-tally# 1. Install dependencies
npm install
# 2. Initialize the database
npm run migrate
# 3. Start using in OpenClaw
# The skill hooks into message-post automatically.
# Use /tasks commands to view analytics.- Local only: All data stays on your machine. No external network calls.
- No message content stored: Only metadata (token count, model, session_id).
- Sandboxed writes: Database is hardcoded to
~/.openclaw/tally/tally.db. - Hook scope: Registered on
message-post— processes every message's metadata only.
A unit of user intent — from a simple question ("What time is it in Tokyo?") to a multi-day campaign ("Scan job market daily"). Every dollar spent is attributed to a task.
TES = quality_score / (normalized_cost × complexity_weight)
Higher is better. A TES > 2.0 means excellent value; < 0.5 means you're overpaying.
- L1 (Reflex): Single-turn Q&A, no tools
- L2 (Routine): Multi-turn or 1–3 tool calls
- L3 (Mission): Multiple tools + file I/O + external APIs
- L4 (Campaign): Sub-agents + cron + cross-session continuity
/tasks list [--level L3] [--status completed]— List recent tasks/tasks stats [--period 30d]— Summary statistics/tasks this-week— This week's task summary/tasks show <task_id>— Task detail view/tasks report --dimension model— Model efficiency breakdown/tasks cron-health— Cron job health check
Three-layer design:
- Task Detector — Identifies task boundaries from message stream
- Task Ledger — Attributes token costs to task IDs (SQLite)
- Analytics Engine — Computes TES and powers dashboards/reports
See PRD.md for the complete product requirements document.
MIT © Jony Jing